Cardiogenic shock is a feared complication of decompensated heart failure (HF). Despite advances in HF therapies, such as revascularization and mechanical circulatory support (MCS), in-hospital mortality for patients with cardiogenic shock remains high, approaching 50%.1,2

Invasive hemodynamic assessment with pulmonary artery catheterization (PAC) allows clinicians to quantify cardiac output, assess intracardiac pressures, differentiate between etiologies of shock, and titrate guideline-directed HF therapies.3 Nevertheless, routine use of PAC in critically ill patients has declined considerably after randomized evidence from other populations—including results of the ESCAPE trial, which studied patients with decompensated HF without cardiogenic shock—failed to show improved outcomes with its use.4,5,6,7,8 The availability of non-invasive modalities of hemodynamic assessment, such as echocardiography, has also contributed to decreasing PAC use.9,10

The role of PAC in cardiogenic, mixed, or undifferentiated shock remains uncertain; these groups may derive the most benefit from invasive hemodynamic assessment to guide therapy. Based on limited evidence, international societies continue to recommend PAC when patients with cardiogenic shock fail to respond to initial medical therapy (Electronic Supplementary Material [ESM] eTable 1).11,12,13,14 We performed a systematic review and meta-analysis of studies evaluating the safety and efficacy—pertaining to mortality and length of stay—of PAC in patients with cardiogenic shock.

Methods

We adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement.15 We pre-registered the protocol with PROSPERO (CRD42019134025).

Study selection

Studies were eligible if: 1) the population consisted of hospitalized adults with a diagnosis of cardiogenic shock as defined by study authors in the spirit of the parameters used in the SHOCK and IABP-SHOCK II trials (see ESM eTable 2);16,17 2) the study design was a randomized controlled trial or comparative observational study; 3) the intervention group received PAC and the comparator group did not; and 4) the outcomes included one or more of the following: mortality (primary outcome), hospital and intensive care unit (ICU) length of stay, and procedural complications (e.g., infection, pneumothorax, arryhthmia, vascular injury).

Search strategy

We searched MEDLINE, CENTRAL, and EMBASE from inception to November 2020, as well as relevant unpublished grey literature including clinical trial registries and major conference proceedings from the past three years (search strategy found in ESM eAppendix). We screened citations of potentially eligible articles without language or publication date restrictions. Two reviewers independently screened titles and abstracts to identify potentially eligible articles for full-text review. Subsequent full-text review was also performed in duplicate. Disagreements between reviewers were resolved by consensus, and if necessary, involvement of a third-party adjudicator.

Data extraction

Two reviewers independently extracted pertinent data from included studies using pre-piloted data collection forms. Disagreements were resolved by consensus. We contacted the corresponding authors if relevant outcome data were missing.

Risk of bias assessment

Two reviewers assessed risk of bias independently and in duplicate using tools developed by the Clinical Advances Through Research and Information Translation (CLARITY) group.18 Studies were judged as having low, unclear, or high risk of bias on the following domains: selection bias, assessment of exposure, assessment of outcomes, adjustment for prognostic variables, assessment for presence/absence of prognostic factors, adequacy of follow-up, and assessment of co-interventions. The overall risk of bias for each included study was categorized as “low” if the risk of bias was low in all domains, “unclear” if the risk of bias was unclear in at least one domain and with no high risk of bias domain, or “high” if the risk of bias was high in at least one domain. Disagreements were resolved by consensus.

Quality of evidence

We used the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) method19 to assess the overall quality of evidence for each outcome.

Statistical analysis

We analyzed data using RevMan software (Review Manager, version 5.3; The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark). We pooled both adjusted and unadjusted results using a random-effects model. Study weights were estimated using the generic inverse variance method20 for adjusted outcomes and the method of DerSimonian and Laird21 for unadjusted outcomes. We report pooled relative risks (RRs) for dichotomous outcomes and mean differences (MDs) for continuous outcomes with corresponding 95% confidence intervals (CIs). We assessed statistical heterogeneity using Chi square and I2 statistics,22 with substantial heterogeneity predefined as P < 0.10 or I2 > 50%.

We inspected funnel plots and performed Egger’s tests to assess for publication bias.23 We explored heterogeneity between studies by performing meta-analyses of predetermined clinically relevant subgroups and comparing effect estimates in RevMan. Subgroups included risk of bias (low, unclear, or high), HF hospitalization status (never hospitalized vs ≥ 1 prior HF hospitalization), cause of cardiogenic shock (acute coronary syndrome, post cardiac surgery, arrhythmia, cardiomyopathy, or not specified), cardiogenic shock post-arrest (or not), and inpatient critical care location (ICU vs cardiac care unit) (ESM eTable 3).

Results

Screening process

Of 833 unique citations, 121 underwent full-text review and 19 studies including ≥ 2,716,287 patients met the eligibility criteria (flow diagram in ESM eFig. 1). Two studies24,25 included in this review did not adequately report breakdown of PAC and no PAC groups and could not be included in the meta-analysis.

Study characteristics

Table 1 describes characteristics of the included studies. Eleven were published peer-reviewed journal articles26,27,28,29,30,31,32,33,34,35,36 and eight were conference proceedings.24,25,37,38,39,40,41,42 We unsuccessfully sought additional data and/or full publications from the authors of two included studies.24,25

Table 1 Summary and characteristics of included studies (19 studies)
Full size table

Fifteen articles were retrospective cohort studies24,26,27,28,29,30,34,35,36,37,38,39,40,41,42 and four were prospective cohort studies;25,31,32,33 no randomized controlled trials met eligibility criteria. One study focused on post cardiac surgery patients39 and seven reported specifically on patients with acute coronary syndromes.26,28,30,34,35,37,40 Length of follow-up varied from index hospitalization up to five years.

Patient characteristics

Nine studies27,28,29,31,32,35,36,38,41 described patient characteristics on admission (Table 2). Patients who received PAC tended to be younger and more often male compared with those that did not. With the exception of one included study,36 comorbidities were similar between groups. Four studies31,32,36,38 reported on etiologies of cardiogenic shock; acute coronary syndrome was the most common.

Table 2 Characteristics of patients on admission (nine studies)
Full size table

Eight studies27,29,31,32,35,36,40,41 described interventions during hospitalization as shown in Table 3. Most patients with cardiogenic shock who underwent revascularization received percutaneous coronary intervention. With the exception of one study,36 patients who received PAC were more likely to receive MCS, vasopressors/inotropes, and mechanical ventilation and require renal replacement therapy.

Table 3 Interventions during hospitalization (eight studies)
Full size table

Risk of bias

We judged risk of bias to be low in two studies,31,33 moderate/unclear in three studies,28,38,39 and high in 14 studies24,25,26,27,29,30,32,34,35,36,37,40,41,42 (see Fig. 1). Domains accounting for the most judgements of high or unclear risk of bias were adjustment for prognostic variables (11 studies) and assessment of co-interventions (17 studies).

Fig. 1
figure 1

Risk of bias summary; review authors’ judgements about each risk of bias item for included studies. Green circles with “+” sign indicate a judgement of low risk of bias. Red circles with “-” sign indicate a judgement of high risk of bias. Yellow circles with “?” sign indicate a judgement of unclear risk of bias.

Full size image

Mortality

We included 17 studies (n = 2,716,206 patients) reporting mortality in our meta-analysis. We evaluated mortality at longest available follow-up, 30-day mortality, and in-hospital mortality. One article reported 28-day mortality;39 we included it in analyses for 30-day mortality.

Pulmonary artery catheterization use was associated with significantly reduced mortality at longest available follow-up for both adjusted (Fig. 2, panel A) and unadjusted results. Nevertheless, there was marked heterogeneity in these results, with all I2 values ≥ 98%. In the adjusted pooled estimate (RR, 0.72; 95% CI, 0.60 to 0.87; I2 = 99%; P = 0.005), eight out of 13 studies27,29,33,35,36,38,40,42 reported in-hospital mortality as the longest follow-up. PAC use was not associated with improved 30-day survival (RR, 0.62; 95% CI, 0.28 to 1.33; I2 = 98%; P = 0.22) but was associated with reduced in-hospital mortality (RR, 0.77; 95% CI, 0.64 to 0.91; I2 = 98%; P = 0.003) (Fig. 2, panel B). The ESM (eFigs 2 to 5) presents charts depicting unadjusted outcomes and 30-day mortality data.

Fig. 2
figure 2

Adjusted mortality at (A) longest available follow-up and (B) in hospital; pulmonary artery catheterization vs no pulmonary artery catheterization. Square markers represent the risk ratio point estimate for each primary study, with size of each square proportional to the weight of the given study in the meta-analysis. Horizontal lines indicate 95% CIs. The solid diamond represents the estimated 95% CI for effect size of all meta-analyzed data. ACS = acute coronary syndrome; CS = cardiogenic shock; PAC = pulmonary artery catheterization.

Full size image

Length of stay

Six studies (n = 1,378,959) reported on hospital length of stay.28,29,32,35,40,41 Pooled unadjusted meta-analysis showed a mean increase of 3.5 days (95% CI, 1.49 to 5.54; I2 = 100%; P = 0.0007) in patients who received PAC (Fig. 3). No studies reported ICU length of stay.

Fig. 3
figure 3

Mean difference in hospital length of stay; pulmonary artery catheterization vs no pulmonary artery catheterization. Square markers represent the point estimate of mean difference in hospital length of stay for each primary study, with size of each square proportional to the weight of the given study in the meta-analysis. Horizontal lines indicate 95% CIs. The solid diamond represents the estimated 95% CI for effect size of all meta-analyzed data.

Full size image

Subgroups and investigation for sources of heterogeneity

Insufficient detail was presented in the included studies to explore the effect of HF hospitalization(s), post-arrest status, or inpatient critical care location on heterogeneity. Analysis by risk of bias did not show a significant difference in effect size across subgroups (ESM, eFig. 6A/B). Although Fig. 2A suggests a subgroup difference pertaining to the etiology of CS, we note that this interpretation is limited by 1) having only one study of patients post cardiac surgery, and 2) the high likelihood that many patients in the “etiology of cardiogenic shock not specified” subgroup having CS due to acute coronary syndromes although not explicitly stated. Accordingly, we felt that our prespecified subgroup analyses did not identify any clear sources of heterogeneity.

Procedural access complications

Data were insufficient to perform a pooled quantitative analysis for procedural access complications. Three studies described complications. In Sionis (2019), seven of 71 patients (9.9%) experienced complications:32 four patients had ventricular tachycardia (VT) during PAC insertion, two of which required cardioversion and three patients experienced minor bleeding. In Rossello (2017), four of 83 patients (4.8%) had major complications attributable to PAC use including heart block, VT, pneumothorax, and catheter-associated bloodstream infection.31 In Sidhu (2017), a higher incidence of pneumothorax was observed in the PAC group; absolute numbers could not be obtained.38

Publication bias

We visually inspected funnel plots for both adjusted and unadjusted outcomes for mortality at longest available follow-up, which included 13 and 15 studies, respectively (see ESM, eFig. 7A/B). We proceeded with Egger’s test as the funnel plots appeared to be asymmetric.23 For adjusted and unadjusted outcomes, we obtained P values of 0.702 and 0.698 respectively, indicating that no publication bias was present.

Quality of evidence

We reviewed the quality of evidence using the GRADE framework (see ESM, eTable 4). All outcomes began as low-quality evidence given the observational nature of data; they were eventually downgraded to very low-quality evidence. All outcomes (hospital length of stay, mortality at longest follow-up, mortality at 30 days, in-hospital mortality) were downgraded for risk of bias due to the greater use of co-interventions during hospitalization among PAC patients. Mortality at 30 days was further downgraded for imprecision because a CI included both benefit and harm of PAC. Although significant heterogeneity (I2 ≥ 90%) was present in our meta-analyses for length of stay and mortality, we elected to not downgrade for serious inconsistency as we felt this was explained by differences in magnitude of effect (large vs small) with overall consistent direction of effect; only one study40 in our review found increased mortality with PAC.

Discussion

In this systematic review and meta-analysis of observational studies, the use of PAC in adults with CS was associated with reduced mortality in hospital and at the longest available follow-up. Thirty-day mortality was not statistically significant, though the direction of effect at this timepoint was consistent. The confidence in our results is limited by the observational nature of included studies, risk of bias, and marked heterogeneity. Furthermore, the quality of evidence for all outcomes was very low.

Despite the PAC population generally being more acutely ill (see Table 3), our adjusted and unadjusted results show an association between PAC use and decreased mortality. Garan et al. (2020) suggest that early complete hemodynamic profiling reduces the need for MCS in the PAC group.36 Our results contrast with existing randomized data of PAC in other populations;4,5,6,7,8 it is plausible that precise knowledge of cardiac indices, vascular resistances, and biventricular filling pressures may allow clinicians to make critical clinical decisions within a narrow therapeutic window to positively impact mortality in patients with CS. Literature indicates the majority of deaths in CS occur early,11,16,17 suggesting PAC may allow for tailored therapy to optimize hemodynamic status, potentially leading to improved outcomes. Because the longest available follow-up in most of our included studies was only during patients’ hospitalization, any benefit shown appears to be driven by reduced in-hospital mortality.

Pulmonary artery catheterization use was associated with longer hospital stay by a mean of 3.5 days. The increased length of hospitalization with PAC could result from more profound CS, more aggressive hemodynamic-guided management, or potentially procedural complications. It may also stem from the higher mortality in patients who did not undergo PAC, leading to shorter hospital stays. Lack of granularity of study-level data did not allow us to explore the timing of death in each group.

Pulmonary artery catheterization is not without harm. In the three studies that reported procedural complications,31,32,38 complications occurred in up to 10% of procedures and ranged from minor bleeding to serious life-threatening adverse events including ventricular dysrhythmias, pneumothorax, and sepsis. These results are consistent with reported complication rates in randomized data from other populations.4,5,6,7Our work has several limitations, the main one being the very low quality of evidence based solely on observational data and the marked heterogeneity of our outcomes. This limits the ability to infer an effect of PAC use on mortality. As PAC is a diagnostic tool that provides data for clinicians to base clinical decisions, it is conceivable that inherent clinical sources of heterogeneity (e.g., provider expertise/training, practitioner/institutional variability, timing of PAC, different definitions of CS) could contribute to the marked heterogeneity seen in our results. Although this may call into question our decision to pool studies, we proceeded with meta-analysis because visual inspection of forest plots showed consistent directions of effect in the majority of included studies and the topic of study remains one where there is clinical equipoise.

Although mean duration of hospitalization was ≤ 30 days in studies reporting hospital length of stay, study-level granularity did not permit comparison of mortality curves across the temporal continuum (in hospital, at 30 days, and at longest available follow-up) between studies. Similarly, as the Society for Cardiovascular Angiography and Intervention only recently updated their classification (stages A to E) of CS,14 it is difficult to determine where along this spectrum of shock our study cohorts fall, making it somewhat challenging to frame in contemporary context.

Lastly, eight of 19 studies24,27,29,35,37,38,40,41 drew from overlapping cohorts of the National Inpatient Sample (NIS). The NIS is the largest database of inpatient hospital stays in the United States, incorporating data from all payers and approximately 20% of American community hospitals.43 A limitation of the NIS is the sheer volume of patient data and with it, potential for marked residual and unmeasured confounding. We cannot exclude the possibility that some patients are being counted more than once; we elected to include all such studies as they were deemed sufficiently distinct in their design and inclusion/exclusion criteria. The observational nature of the pooled studies in this review further increases the risk of confounding by indication; as such, our results should be interpreted with caution.

Contemporary recommendations11,12,13,14 for PAC use in CS are based on individual observational studies and expert consensus. Our meta-analysis provides the most comprehensive and rigorous analysis of PAC in CS published to date, following a prespecified protocol and using high methodological standards.15 In patients who do not have a separate indication for invasive hemodynamic assessment (e.g., those being worked up for advanced therapies), prospective randomized clinical trials are needed to further characterize the role of PAC in patients presenting with CS. In addition, further work should assess the cost-effectiveness of routine PAC in CS, study the optimal timing of PAC, and/or seek to understand practitioner-to-practitioner variation in decision-making based on hemodynamic values.

Conclusions

In this systematic review and meta-analysis of observational studies, our work suggests that PAC use in patients with CS is associated with lower mortality. The observed increase in hospital length of stay may represent survivor bias or relate to more aggressive management with PAC. Overall, these results support consideration of PAC for hemodynamic assessment in CS; however, our confidence in this conclusion is diminished by the very low quality of available evidence and marked heterogeneity of included studies.